Bias, checked.
Also called: Cognitive bias audit · Red-teaming your idea · Disconfirmation check · Devil’s-advocate review
Deliberately turning your evaluation on your own thinking, hunting the comfortable lies that make a weak idea look strong to its founder.
You are the least reliable judge of your own idea, because you want it to be true. Confirmation bias, sunk cost and optimism quietly rig every evaluation in your favour. The fix is structural, not willpower: design a check that hunts disconfirming evidence on purpose.
What checking bias is
Here is the uncomfortable claim this whole activity rests on: the more you love an idea, the worse your judgement of it becomes, and you will not feel it happening. Bias does not announce itself. It feels exactly like clear thinking. That is what makes it dangerous, and that is why you cannot fix it by simply resolving to be objective.
Three biases do most of the damage at this stage. Confirmation bias makes you notice the customers who say yes and quietly file the ones who say no under “not my target market”. Sunk-cost makes the months you have already spent feel like a reason to continue rather than money already gone. Optimism and the planning fallacy make you cost the cheap version, the fast timeline, the world where nothing goes wrong. Each one is a lie you tell yourself because the truth is less pleasant.
Checking bias means accepting you cannot out-think your own brain and building a process that does the work instead. You do not ask “is my idea good?” You design a check whose explicit job is to find evidence you are wrong, and you go looking for it as hard as you went looking for reasons to start.
How to design the check
- Appoint a devil’s advocate. Give one person the formal job of arguing the idea should be killed. Not “play nice and raise concerns”, but build the strongest possible case against. A real role, not a vibe.
- Run a pre-mortem. Sit the team in an imagined future where the launch has flopped, and ask everyone to write down why. People will say things looking backwards from failure that they would never volunteer looking forwards from hope.
- Go and find the disconfirming evidence. Deliberately seek the customer who is happy without you, the cheaper workaround, the reason this has not already been built. If you cannot find any, you have not looked hard enough.
How checking bias played out
Dan and Anna loved the proofing box. That was the problem. Because they loved it, every conversation seemed to confirm it, and the signals that should have given them pause slid past unnoticed. Here is the bias that was actually at work on the project, and the check we built to counter each one.
The biases did not go away. We did not make Dan and Anna more objective, because you cannot. What we did was put a process between them and the decision, and the process did the doubting they could not.
- “Everyone I asked loved it.”
- Counting the yeses, filing the noes.
- “We’ve come too far to stop now.”
- Costing the version where nothing goes wrong.
- “Who is happy without this, and why?”
- Counting the noes as hard as the yeses.
- “What would past spend tell a fresh team?”
- Costing the version where the build runs long.
The test is simple. If your evaluation only ever produces reasons to keep going, it is not an evaluation. It is a sales pitch you are giving yourself.
How it fits the bigger picture
Check Bias is activity 04.10.04 in the framework, sitting inside Stage 04 Evaluate. It runs over the top of your risk assessment and competitor work, then hands a sounder, stress-tested view forward into run workshops (04.10.05), where the team turns evidence into shared decisions.
What it can do
It catches the comfortable lies before they cost you money. A bias check turns a private hunch you are quietly defending into a claim the whole team has tried, and failed, to knock down. What survives a real attempt to disprove it is worth far more than what survived no scrutiny at all.
What it can’t do
It can’t make you objective; nothing can. It only buys you a process that doubts when you won’t. And it can’t manufacture evidence: if you run a half-hearted devil’s advocate who agrees with you, you have performed the check without doing it, which is worse than skipping it, because now you feel safe.
See the full 10-stage process →
Try it yourself
Take the idea you are most confident about. Write down, in one sentence, what would have to be true for it to fail, then spend an honest hour trying to prove that sentence right rather than wrong. Talk to someone with no stake in the outcome. If the exercise feels uncomfortable, you are finally doing it properly.
Or run the guided version, where the GPT will press the assumptions you would rather not look at. Start the Free Sprint →
Your bias-check checklist
Project notes: the bakers who didn’t buy
▸ From the notebook · optional reading
Dan and Anna Hartley loved the proofing box too much to judge it. So in Stockport, we made them go and find the people who didn’t want it.
3 min read · click to open
By the time the idea reached the bias check, Dan and Anna were in love with it, and that love was doing their thinking for them. Every baker they mentioned it to was enthusiastic, which they took as the market shouting yes. I had to point out that they had only ever asked the people most likely to agree.
The check we built
I asked them to do one uncomfortable thing: go and find the bakers who were perfectly happy without it. Not the lukewarm ones, the genuinely content ones, the people who put the dough in the cooling oven under the light and got a fine loaf, year after year, for nothing.
They found them quickly, and the conversations stung. A surprising number of serious bakers had a free, good-enough method and no intention of paying £149 to replace it. That was the disconfirming evidence the love had been hiding.
What it changed
It did not kill the box. It sharpened the target. The product was never for the happy oven-light baker; it was for the one whose cold kitchen made the oven-light method unreliable through a UK winter. The check did not give us a no. It gave us a much narrower, much truer yes, and it stopped us pricing a build where the UKCA and BS EN 61010 route ran perfectly to plan, which it never does.
— Evaluate stage, project notes, 2026
— Next in Evaluate → Run workshops